<![CDATA[An accurate sales forecasting is crucial to the profits earned because it affects the company's stock management. With computational support, machine learning and artificial intelligence can continuously and automatically recognize patterns in data, thereby reducing the risk of demand unpredictability. PT XYZ is one of the companies in industrial sector that produces various beverage products. The factory in Medan. One of the products is the tea glass bottle. At PT XYZ, there are frequent differences between forecasting data and sales data, causing high error rates in production planning accuracy. This study aims to analyze the most effective model for forecasting future sales by comparing the accuracy of a Machine Learning-based forecasting model with the existing forecasting method currently employed at PT XYZ. This research was conducted using the Recurrent Neural network (RNN) method as part of the Machine Learning approach. The data that was inputted to the programme was weekly demand data, calendar day off data, temperature data, and population data. The forecasted data is weekly demand. Based on the company's historical data, a demand graph is obtained which has a cyclical pattern. From the results of forecasting using Machine Learning, an accuracy value of 99.47% is obtained with an error rate of 0.53%, which is still below the tolerance limit set by the company. The error rate shows a decrease of 14.72% compared to the error value in the previous company model. This decrease is expected to help control inventory more effectively. ]]>
